Study design, setting and participants
This post hoc analysis of a nationwide multicenter retrospective cohort study (Japan Septic Disseminated Intravascular Coagulation; J-SEPTIC DIC study) was conducted in 42 intensive care units (ICUs) in Japan.8 All adult patients diagnosed as having severe sepsis or septic shock by Sepsis-1 criteria during January 2011 through December 2013 were consecutively enrolled in the registry. We also included patients with septic DIC based on the Sepsis-3 criteria and Japanese Association for Acute Medicine (JAAM) DIC criteria. Exclusion criteria included the use of warfarin/acetylsalicylic acid/thrombolytic therapies before study entry; the limitation of sustained life care or post-cardiopulmonary arrest resuscitation status; history of fulminant hepatitis, decompensated liver cirrhosis, or other serious liver disorder; treatment with any chemotherapy; history of hematologic malignant disease; other conditions increasing thrombosis risk at study entry; and patients with missing data for main analysis. We defined the day of study inclusion as “day 1”. This study followed the principles of the Declaration of Helsinki. The study protocol was reviewed by the Ethics Committees of Osaka University Hospital, and the board waived the need for the registration because this study used an open and anonymous dataset.
Definition of sepsis and DIC
We diagnosed “sepsis” according to the Sepsis-3 definition presented at the 45th Critical Care Congress of the Society of Critical Care Medicine in 2016.9 In this new definition, sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection. Septic shock was also defined based on Sepsis-3 criteria as an elevated lactate level and sepsis-induced hypotension persisting despite adequate fluid resuscitation and requiring catecholamine infusion to improve hemodynamic status.
DIC was diagnosed at the time of inclusion based on JAAM DIC criteria. The JAAM DIC scoring system uses the SIRS score and global coagulation tests including platelet counts, prothrombin time, and FDP/D-dimer levels.10 We also evaluated the ISTH overt DIC score proposed by the Scientific Subcommittee on DIC of the ISTH.11 To calculate this score, FDP values were adopted as the fibrin-related marker and scored according to cut-off levels and ranges previously published by Gando et al.12
Data collection
Patients were followed until hospital discharge or death. A case report form was developed for the J-SEPTIC DIC registry, and the following information was obtained: age, sex, illness severity scores on the day of ICU admission, source of ICU admission, preexisting comorbidities, primary source of infection, and therapeutic interventions such as immunoglobulin, low-dose steroid, renal replacement therapy, and low-dose heparin for prophylaxis against deep vein thrombosis (DVT). Illness severity was evaluated according to the Sequential Organ Failure Assessment (SOFA) score evaluated on days 1 and 4 and the Acute Physiology and Chronic Health Evaluation (APACHE) II score evaluated on day 1.
The primary outcome measure was all-cause in-hospital mortality. Secondary outcomes were severe bleeding events, which included the occurrence of intracranial hemorrhage, transfusion requirements related to bleeding, and bleeding requiring surgical intervention.
Patient categorization
Study patients were categorized into the anticoagulant group (those who underwent systemic administration at therapeutic doses of any anticoagulant agents such as antithrombin, recombinant human thrombomodulin, heparin/heparinoid or serine protease inhibitors) and the control group (who received no anticoagulant therapies for the purpose of DIC treatment). Patients receiving prophylactic administration of low-dose heparin/heparinoid for venous thromboembolism were included in both groups.
Statistical analysis
To evaluate differences in efficacy and safety of anticoagulant therapies according to patient age, we conducted logistic regression analysis for primary and secondary outcomes including interactions terms between the treatment variable and patient age.
We also stratified the participants into three subsets based on their age: age ≤ 60, age 60–79, and age ≥ 80 years. We conducted multivariable Cox proportional hazard regression analyses to examine the effect of anticoagulant therapies on mortality in each age class separately.
The retrospective design of this study caused baseline imbalances between the anticoagulant and control groups; therefore, all regression models were adjusted using propensity scores. The propensity score for the likelihood of undergoing anticoagulant therapies that patients actually received was calculated using multivariable logistic regression analysis including sex, pre-existing comorbidities, disease severity scores, primary source of infection, causal microorganisms, and other therapeutic interventions as covariates. The detailed combinations of the variables are described in Supplementary Table S1.
Descriptive statistics were calculated as medians (interquartile range) or proportions (numbers), as appropriate. Univariate differences between groups were assessed with the Mann-Whitney U test or chi-square test, as appropriate. Missing values were not imputed in any of the regression models.
All statistical inferences were performed with a 2-sided p at 5% significance level. Because of the underpowered nature of the interaction analysis, we used a 2-sided significance level of 20% with statistical inferences for the interaction analyses.13 All statistical analyses were conducted using STATA Data Analysis and Statistical Software version 15.0 (StataCorp, College Station, TX).